1 /*
2 * Copyright (c) 1997, 2025, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #ifndef CPU_X86_MACROASSEMBLER_X86_HPP
26 #define CPU_X86_MACROASSEMBLER_X86_HPP
27
28 #include "asm/assembler.hpp"
29 #include "asm/register.hpp"
30 #include "code/vmreg.inline.hpp"
31 #include "compiler/oopMap.hpp"
32 #include "utilities/macros.hpp"
33 #include "runtime/signature.hpp"
34 #include "runtime/vm_version.hpp"
35 #include "utilities/checkedCast.hpp"
36
37 class ciInlineKlass;
38
39 // MacroAssembler extends Assembler by frequently used macros.
40 //
41 // Instructions for which a 'better' code sequence exists depending
42 // on arguments should also go in here.
43
44 class MacroAssembler: public Assembler {
45 friend class LIR_Assembler;
46 friend class Runtime1; // as_Address()
47
48 public:
49 // Support for VM calls
50 //
51 // This is the base routine called by the different versions of call_VM_leaf. The interpreter
52 // may customize this version by overriding it for its purposes (e.g., to save/restore
53 // additional registers when doing a VM call).
54
55 virtual void call_VM_leaf_base(
56 address entry_point, // the entry point
57 int number_of_arguments // the number of arguments to pop after the call
58 );
59
60 protected:
61 // This is the base routine called by the different versions of call_VM. The interpreter
62 // may customize this version by overriding it for its purposes (e.g., to save/restore
63 // additional registers when doing a VM call).
64 //
65 // call_VM_base returns the register which contains the thread upon return.
66 // If no last_java_sp is specified (noreg) than rsp will be used instead.
67 virtual void call_VM_base( // returns the register containing the thread upon return
68 Register oop_result, // where an oop-result ends up if any; use noreg otherwise
69 Register last_java_sp, // to set up last_Java_frame in stubs; use noreg otherwise
70 address entry_point, // the entry point
71 int number_of_arguments, // the number of arguments (w/o thread) to pop after the call
72 bool check_exceptions // whether to check for pending exceptions after return
73 );
74
75 void call_VM_helper(Register oop_result, address entry_point, int number_of_arguments, bool check_exceptions = true);
76
77 public:
78 MacroAssembler(CodeBuffer* code) : Assembler(code) {}
79
80 // These routines should emit JVMTI PopFrame and ForceEarlyReturn handling code.
81 // The implementation is only non-empty for the InterpreterMacroAssembler,
82 // as only the interpreter handles PopFrame and ForceEarlyReturn requests.
83 virtual void check_and_handle_popframe();
84 virtual void check_and_handle_earlyret();
85
86 Address as_Address(AddressLiteral adr);
87 Address as_Address(ArrayAddress adr, Register rscratch);
88
89 // Support for null-checks
90 //
91 // Generates code that causes a null OS exception if the content of reg is null.
92 // If the accessed location is M[reg + offset] and the offset is known, provide the
93 // offset. No explicit code generation is needed if the offset is within a certain
94 // range (0 <= offset <= page_size).
95
96 void null_check(Register reg, int offset = -1);
97 static bool needs_explicit_null_check(intptr_t offset);
98 static bool uses_implicit_null_check(void* address);
99
100 // markWord tests, kills markWord reg
101 void test_markword_is_inline_type(Register markword, Label& is_inline_type);
102
103 // inlineKlass queries, kills temp_reg
104 void test_oop_is_not_inline_type(Register object, Register tmp, Label& not_inline_type, bool can_be_null = true);
105
106 void test_field_is_null_free_inline_type(Register flags, Register temp_reg, Label& is_null_free);
107 void test_field_is_not_null_free_inline_type(Register flags, Register temp_reg, Label& not_null_free);
108 void test_field_is_flat(Register flags, Register temp_reg, Label& is_flat);
109 void test_field_has_null_marker(Register flags, Register temp_reg, Label& has_null_marker);
110
111 // Check oops for special arrays, i.e. flat arrays and/or null-free arrays
112 void test_oop_prototype_bit(Register oop, Register temp_reg, int32_t test_bit, bool jmp_set, Label& jmp_label);
113 void test_flat_array_oop(Register oop, Register temp_reg, Label& is_flat_array);
114 void test_non_flat_array_oop(Register oop, Register temp_reg, Label& is_non_flat_array);
115 void test_null_free_array_oop(Register oop, Register temp_reg, Label& is_null_free_array);
116 void test_non_null_free_array_oop(Register oop, Register temp_reg, Label& is_non_null_free_array);
117
118 // Check array klass layout helper for flat or null-free arrays...
119 void test_flat_array_layout(Register lh, Label& is_flat_array);
120 void test_non_flat_array_layout(Register lh, Label& is_non_flat_array);
121
122 // Required platform-specific helpers for Label::patch_instructions.
123 // They _shadow_ the declarations in AbstractAssembler, which are undefined.
124 void pd_patch_instruction(address branch, address target, const char* file, int line) {
125 unsigned char op = branch[0];
126 assert(op == 0xE8 /* call */ ||
127 op == 0xE9 /* jmp */ ||
128 op == 0xEB /* short jmp */ ||
129 (op & 0xF0) == 0x70 /* short jcc */ ||
130 (op == 0x0F && (branch[1] & 0xF0) == 0x80) /* jcc */ ||
131 (op == 0xC7 && branch[1] == 0xF8) /* xbegin */ ||
132 (op == 0x8D) /* lea */,
133 "Invalid opcode at patch point");
134
135 if (op == 0xEB || (op & 0xF0) == 0x70) {
136 // short offset operators (jmp and jcc)
137 char* disp = (char*) &branch[1];
138 int imm8 = checked_cast<int>(target - (address) &disp[1]);
139 guarantee(this->is8bit(imm8), "Short forward jump exceeds 8-bit offset at %s:%d",
140 file == nullptr ? "<null>" : file, line);
141 *disp = (char)imm8;
142 } else {
143 int* disp = (int*) &branch[(op == 0x0F || op == 0xC7 || op == 0x8D) ? 2 : 1];
144 int imm32 = checked_cast<int>(target - (address) &disp[1]);
145 *disp = imm32;
146 }
147 }
148
149 // The following 4 methods return the offset of the appropriate move instruction
150
151 // Support for fast byte/short loading with zero extension (depending on particular CPU)
152 int load_unsigned_byte(Register dst, Address src);
153 int load_unsigned_short(Register dst, Address src);
154
155 // Support for fast byte/short loading with sign extension (depending on particular CPU)
156 int load_signed_byte(Register dst, Address src);
157 int load_signed_short(Register dst, Address src);
158
159 // Support for sign-extension (hi:lo = extend_sign(lo))
160 void extend_sign(Register hi, Register lo);
161
162 // Load and store values by size and signed-ness
163 void load_sized_value(Register dst, Address src, size_t size_in_bytes, bool is_signed, Register dst2 = noreg);
164 void store_sized_value(Address dst, Register src, size_t size_in_bytes, Register src2 = noreg);
165
166 // Support for inc/dec with optimal instruction selection depending on value
167
168 void increment(Register reg, int value = 1) { incrementq(reg, value); }
169 void decrement(Register reg, int value = 1) { decrementq(reg, value); }
170 void increment(Address dst, int value = 1) { incrementq(dst, value); }
171 void decrement(Address dst, int value = 1) { decrementq(dst, value); }
172
173 void decrementl(Address dst, int value = 1);
174 void decrementl(Register reg, int value = 1);
175
176 void decrementq(Register reg, int value = 1);
177 void decrementq(Address dst, int value = 1);
178
179 void incrementl(Address dst, int value = 1);
180 void incrementl(Register reg, int value = 1);
181
182 void incrementq(Register reg, int value = 1);
183 void incrementq(Address dst, int value = 1);
184
185 void incrementl(AddressLiteral dst, Register rscratch = noreg);
186 void incrementl(ArrayAddress dst, Register rscratch);
187
188 void incrementq(AddressLiteral dst, Register rscratch = noreg);
189
190 // Support optimal SSE move instructions.
191 void movflt(XMMRegister dst, XMMRegister src) {
192 if (dst-> encoding() == src->encoding()) return;
193 if (UseXmmRegToRegMoveAll) { movaps(dst, src); return; }
194 else { movss (dst, src); return; }
195 }
196 void movflt(XMMRegister dst, Address src) { movss(dst, src); }
197 void movflt(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
198 void movflt(Address dst, XMMRegister src) { movss(dst, src); }
199
200 // Move with zero extension
201 void movfltz(XMMRegister dst, XMMRegister src) { movss(dst, src); }
202
203 void movdbl(XMMRegister dst, XMMRegister src) {
204 if (dst-> encoding() == src->encoding()) return;
205 if (UseXmmRegToRegMoveAll) { movapd(dst, src); return; }
206 else { movsd (dst, src); return; }
207 }
208
209 void movdbl(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
210
211 void movdbl(XMMRegister dst, Address src) {
212 if (UseXmmLoadAndClearUpper) { movsd (dst, src); return; }
213 else { movlpd(dst, src); return; }
214 }
215 void movdbl(Address dst, XMMRegister src) { movsd(dst, src); }
216
217 void flt_to_flt16(Register dst, XMMRegister src, XMMRegister tmp) {
218 // Use separate tmp XMM register because caller may
219 // requires src XMM register to be unchanged (as in x86.ad).
220 vcvtps2ph(tmp, src, 0x04, Assembler::AVX_128bit);
221 movdl(dst, tmp);
222 movswl(dst, dst);
223 }
224
225 void flt16_to_flt(XMMRegister dst, Register src) {
226 movdl(dst, src);
227 vcvtph2ps(dst, dst, Assembler::AVX_128bit);
228 }
229
230 // Alignment
231 void align32();
232 void align64();
233 void align(uint modulus);
234 void align(uint modulus, uint target);
235
236 void post_call_nop();
237
238 // Stack frame creation/removal
239 void enter();
240 void leave();
241
242 // Support for getting the JavaThread pointer (i.e.; a reference to thread-local information).
243 // The pointer will be loaded into the thread register. This is a slow version that does native call.
244 // Normally, JavaThread pointer is available in r15_thread, use that where possible.
245 void get_thread_slow(Register thread);
246
247 // Support for argument shuffling
248
249 // bias in bytes
250 void move32_64(VMRegPair src, VMRegPair dst, Register tmp = rax, int in_stk_bias = 0, int out_stk_bias = 0);
251 void long_move(VMRegPair src, VMRegPair dst, Register tmp = rax, int in_stk_bias = 0, int out_stk_bias = 0);
252 void float_move(VMRegPair src, VMRegPair dst, Register tmp = rax, int in_stk_bias = 0, int out_stk_bias = 0);
253 void double_move(VMRegPair src, VMRegPair dst, Register tmp = rax, int in_stk_bias = 0, int out_stk_bias = 0);
254 void move_ptr(VMRegPair src, VMRegPair dst);
255 void object_move(OopMap* map,
256 int oop_handle_offset,
257 int framesize_in_slots,
258 VMRegPair src,
259 VMRegPair dst,
260 bool is_receiver,
261 int* receiver_offset);
262
263 // Support for VM calls
264 //
265 // It is imperative that all calls into the VM are handled via the call_VM macros.
266 // They make sure that the stack linkage is setup correctly. call_VM's correspond
267 // to ENTRY/ENTRY_X entry points while call_VM_leaf's correspond to LEAF entry points.
268
269
270 void call_VM(Register oop_result,
271 address entry_point,
272 bool check_exceptions = true);
273 void call_VM(Register oop_result,
274 address entry_point,
275 Register arg_1,
276 bool check_exceptions = true);
277 void call_VM(Register oop_result,
278 address entry_point,
279 Register arg_1, Register arg_2,
280 bool check_exceptions = true);
281 void call_VM(Register oop_result,
282 address entry_point,
283 Register arg_1, Register arg_2, Register arg_3,
284 bool check_exceptions = true);
285
286 // Overloadings with last_Java_sp
287 void call_VM(Register oop_result,
288 Register last_java_sp,
289 address entry_point,
290 int number_of_arguments = 0,
291 bool check_exceptions = true);
292 void call_VM(Register oop_result,
293 Register last_java_sp,
294 address entry_point,
295 Register arg_1, bool
296 check_exceptions = true);
297 void call_VM(Register oop_result,
298 Register last_java_sp,
299 address entry_point,
300 Register arg_1, Register arg_2,
301 bool check_exceptions = true);
302 void call_VM(Register oop_result,
303 Register last_java_sp,
304 address entry_point,
305 Register arg_1, Register arg_2, Register arg_3,
306 bool check_exceptions = true);
307
308 void get_vm_result_oop(Register oop_result);
309 void get_vm_result_metadata(Register metadata_result);
310
311 // These always tightly bind to MacroAssembler::call_VM_base
312 // bypassing the virtual implementation
313 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, int number_of_arguments = 0, bool check_exceptions = true);
314 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, bool check_exceptions = true);
315 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, bool check_exceptions = true);
316 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions = true);
317 void super_call_VM(Register oop_result, Register last_java_sp, address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4, bool check_exceptions = true);
318
319 void call_VM_leaf0(address entry_point);
320 void call_VM_leaf(address entry_point,
321 int number_of_arguments = 0);
322 void call_VM_leaf(address entry_point,
323 Register arg_1);
324 void call_VM_leaf(address entry_point,
325 Register arg_1, Register arg_2);
326 void call_VM_leaf(address entry_point,
327 Register arg_1, Register arg_2, Register arg_3);
328
329 void call_VM_leaf(address entry_point,
330 Register arg_1, Register arg_2, Register arg_3, Register arg_4);
331
332 // These always tightly bind to MacroAssembler::call_VM_leaf_base
333 // bypassing the virtual implementation
334 void super_call_VM_leaf(address entry_point);
335 void super_call_VM_leaf(address entry_point, Register arg_1);
336 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2);
337 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3);
338 void super_call_VM_leaf(address entry_point, Register arg_1, Register arg_2, Register arg_3, Register arg_4);
339
340 void set_last_Java_frame(Register last_java_sp,
341 Register last_java_fp,
342 address last_java_pc,
343 Register rscratch);
344
345 void set_last_Java_frame(Register last_java_sp,
346 Register last_java_fp,
347 Label &last_java_pc,
348 Register scratch);
349
350 void reset_last_Java_frame(bool clear_fp);
351
352 // jobjects
353 void clear_jobject_tag(Register possibly_non_local);
354 void resolve_jobject(Register value, Register tmp);
355 void resolve_global_jobject(Register value, Register tmp);
356
357 // C 'boolean' to Java boolean: x == 0 ? 0 : 1
358 void c2bool(Register x);
359
360 // C++ bool manipulation
361
362 void movbool(Register dst, Address src);
363 void movbool(Address dst, bool boolconst);
364 void movbool(Address dst, Register src);
365 void testbool(Register dst);
366
367 void resolve_oop_handle(Register result, Register tmp);
368 void resolve_weak_handle(Register result, Register tmp);
369 void load_mirror(Register mirror, Register method, Register tmp);
370 void load_method_holder_cld(Register rresult, Register rmethod);
371
372 void load_method_holder(Register holder, Register method);
373
374 // oop manipulations
375
376 // Load oopDesc._metadata without decode (useful for direct Klass* compare from oops)
377 void load_metadata(Register dst, Register src);
378 void load_narrow_klass_compact(Register dst, Register src);
379 void load_klass(Register dst, Register src, Register tmp);
380 void store_klass(Register dst, Register src, Register tmp);
381
382 // Compares the Klass pointer of an object to a given Klass (which might be narrow,
383 // depending on UseCompressedClassPointers).
384 void cmp_klass(Register klass, Register obj, Register tmp);
385
386 // Compares the Klass pointer of two objects obj1 and obj2. Result is in the condition flags.
387 // Uses tmp1 and tmp2 as temporary registers.
388 void cmp_klasses_from_objects(Register obj1, Register obj2, Register tmp1, Register tmp2);
389
390 void access_load_at(BasicType type, DecoratorSet decorators, Register dst, Address src,
391 Register tmp1);
392 void access_store_at(BasicType type, DecoratorSet decorators, Address dst, Register val,
393 Register tmp1, Register tmp2, Register tmp3);
394
395 void flat_field_copy(DecoratorSet decorators, Register src, Register dst, Register inline_layout_info);
396
397 // inline type data payload offsets...
398 void payload_offset(Register inline_klass, Register offset);
399 void payload_addr(Register oop, Register data, Register inline_klass);
400 // get data payload ptr a flat value array at index, kills rcx and index
401 void data_for_value_array_index(Register array, Register array_klass,
402 Register index, Register data);
403
404 void load_heap_oop(Register dst, Address src, Register tmp1 = noreg, DecoratorSet decorators = 0);
405 void load_heap_oop_not_null(Register dst, Address src, Register tmp1 = noreg, DecoratorSet decorators = 0);
406 void store_heap_oop(Address dst, Register val, Register tmp1 = noreg,
407 Register tmp2 = noreg, Register tmp3 = noreg, DecoratorSet decorators = 0);
408
409 // Used for storing null. All other oop constants should be
410 // stored using routines that take a jobject.
411 void store_heap_oop_null(Address dst);
412
413 void load_prototype_header(Register dst, Register src, Register tmp);
414
415 void store_klass_gap(Register dst, Register src);
416
417 // This dummy is to prevent a call to store_heap_oop from
418 // converting a zero (like null) into a Register by giving
419 // the compiler two choices it can't resolve
420
421 void store_heap_oop(Address dst, void* dummy);
422
423 void encode_heap_oop(Register r);
424 void decode_heap_oop(Register r);
425 void encode_heap_oop_not_null(Register r);
426 void decode_heap_oop_not_null(Register r);
427 void encode_heap_oop_not_null(Register dst, Register src);
428 void decode_heap_oop_not_null(Register dst, Register src);
429
430 void set_narrow_oop(Register dst, jobject obj);
431 void set_narrow_oop(Address dst, jobject obj);
432 void cmp_narrow_oop(Register dst, jobject obj);
433 void cmp_narrow_oop(Address dst, jobject obj);
434
435 void encode_klass_not_null(Register r, Register tmp);
436 void decode_klass_not_null(Register r, Register tmp);
437 void encode_and_move_klass_not_null(Register dst, Register src);
438 void decode_and_move_klass_not_null(Register dst, Register src);
439 void set_narrow_klass(Register dst, Klass* k);
440 void set_narrow_klass(Address dst, Klass* k);
441 void cmp_narrow_klass(Register dst, Klass* k);
442 void cmp_narrow_klass(Address dst, Klass* k);
443
444 // if heap base register is used - reinit it with the correct value
445 void reinit_heapbase();
446
447 DEBUG_ONLY(void verify_heapbase(const char* msg);)
448
449 // Int division/remainder for Java
450 // (as idivl, but checks for special case as described in JVM spec.)
451 // returns idivl instruction offset for implicit exception handling
452 int corrected_idivl(Register reg);
453
454 // Long division/remainder for Java
455 // (as idivq, but checks for special case as described in JVM spec.)
456 // returns idivq instruction offset for implicit exception handling
457 int corrected_idivq(Register reg);
458
459 void int3();
460
461 // Long operation macros for a 32bit cpu
462 // Long negation for Java
463 void lneg(Register hi, Register lo);
464
465 // Long multiplication for Java
466 // (destroys contents of eax, ebx, ecx and edx)
467 void lmul(int x_rsp_offset, int y_rsp_offset); // rdx:rax = x * y
468
469 // Long shifts for Java
470 // (semantics as described in JVM spec.)
471 void lshl(Register hi, Register lo); // hi:lo << (rcx & 0x3f)
472 void lshr(Register hi, Register lo, bool sign_extension = false); // hi:lo >> (rcx & 0x3f)
473
474 // Long compare for Java
475 // (semantics as described in JVM spec.)
476 void lcmp2int(Register x_hi, Register x_lo, Register y_hi, Register y_lo); // x_hi = lcmp(x, y)
477
478
479 // misc
480
481 // Sign extension
482 void sign_extend_short(Register reg);
483 void sign_extend_byte(Register reg);
484
485 // Division by power of 2, rounding towards 0
486 void division_with_shift(Register reg, int shift_value);
487
488 // dst = c = a * b + c
489 void fmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c);
490 void fmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c);
491
492 void vfmad(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len);
493 void vfmaf(XMMRegister dst, XMMRegister a, XMMRegister b, XMMRegister c, int vector_len);
494 void vfmad(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len);
495 void vfmaf(XMMRegister dst, XMMRegister a, Address b, XMMRegister c, int vector_len);
496
497
498 // same as fcmp2int, but using SSE2
499 void cmpss2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less);
500 void cmpsd2int(XMMRegister opr1, XMMRegister opr2, Register dst, bool unordered_is_less);
501
502 void push_IU_state();
503 void pop_IU_state();
504
505 void push_FPU_state();
506 void pop_FPU_state();
507
508 void push_CPU_state();
509 void pop_CPU_state();
510
511 void push_cont_fastpath();
512 void pop_cont_fastpath();
513
514 DEBUG_ONLY(void stop_if_in_cont(Register cont_reg, const char* name);)
515
516 // Round up to a power of two
517 void round_to(Register reg, int modulus);
518
519 private:
520 // General purpose and XMM registers potentially clobbered by native code; there
521 // is no need for FPU or AVX opmask related methods because C1/interpreter
522 // - we save/restore FPU state as a whole always
523 // - do not care about AVX-512 opmask
524 static RegSet call_clobbered_gp_registers();
525 static XMMRegSet call_clobbered_xmm_registers();
526
527 void push_set(XMMRegSet set, int offset);
528 void pop_set(XMMRegSet set, int offset);
529
530 public:
531 void push_set(RegSet set, int offset = -1);
532 void pop_set(RegSet set, int offset = -1);
533
534 // Push and pop everything that might be clobbered by a native
535 // runtime call.
536 // Only save the lower 64 bits of each vector register.
537 // Additional registers can be excluded in a passed RegSet.
538 void push_call_clobbered_registers_except(RegSet exclude, bool save_fpu = true);
539 void pop_call_clobbered_registers_except(RegSet exclude, bool restore_fpu = true);
540
541 void push_call_clobbered_registers(bool save_fpu = true) {
542 push_call_clobbered_registers_except(RegSet(), save_fpu);
543 }
544 void pop_call_clobbered_registers(bool restore_fpu = true) {
545 pop_call_clobbered_registers_except(RegSet(), restore_fpu);
546 }
547
548 // allocation
549
550 // Object / value buffer allocation...
551 // Allocate instance of klass, assumes klass initialized by caller
552 // new_obj prefers to be rax
553 // Kills t1 and t2, perserves klass, return allocation in new_obj (rsi on LP64)
554 void allocate_instance(Register klass, Register new_obj,
555 Register t1, Register t2,
556 bool clear_fields, Label& alloc_failed);
557
558 void tlab_allocate(
559 Register obj, // result: pointer to object after successful allocation
560 Register var_size_in_bytes, // object size in bytes if unknown at compile time; invalid otherwise
561 int con_size_in_bytes, // object size in bytes if known at compile time
562 Register t1, // temp register
563 Register t2, // temp register
564 Label& slow_case // continuation point if fast allocation fails
565 );
566 void zero_memory(Register address, Register length_in_bytes, int offset_in_bytes, Register temp);
567
568 // For field "index" within "klass", return inline_klass ...
569 void get_inline_type_field_klass(Register klass, Register index, Register inline_klass);
570
571 void inline_layout_info(Register klass, Register index, Register layout_info);
572
573 void population_count(Register dst, Register src, Register scratch1, Register scratch2);
574
575 // interface method calling
576 void lookup_interface_method(Register recv_klass,
577 Register intf_klass,
578 RegisterOrConstant itable_index,
579 Register method_result,
580 Register scan_temp,
581 Label& no_such_interface,
582 bool return_method = true);
583
584 void lookup_interface_method_stub(Register recv_klass,
585 Register holder_klass,
586 Register resolved_klass,
587 Register method_result,
588 Register scan_temp,
589 Register temp_reg2,
590 Register receiver,
591 int itable_index,
592 Label& L_no_such_interface);
593
594 // virtual method calling
595 void lookup_virtual_method(Register recv_klass,
596 RegisterOrConstant vtable_index,
597 Register method_result);
598
599 // Test sub_klass against super_klass, with fast and slow paths.
600
601 // The fast path produces a tri-state answer: yes / no / maybe-slow.
602 // One of the three labels can be null, meaning take the fall-through.
603 // If super_check_offset is -1, the value is loaded up from super_klass.
604 // No registers are killed, except temp_reg.
605 void check_klass_subtype_fast_path(Register sub_klass,
606 Register super_klass,
607 Register temp_reg,
608 Label* L_success,
609 Label* L_failure,
610 Label* L_slow_path,
611 RegisterOrConstant super_check_offset = RegisterOrConstant(-1));
612
613 // The rest of the type check; must be wired to a corresponding fast path.
614 // It does not repeat the fast path logic, so don't use it standalone.
615 // The temp_reg and temp2_reg can be noreg, if no temps are available.
616 // Updates the sub's secondary super cache as necessary.
617 // If set_cond_codes, condition codes will be Z on success, NZ on failure.
618 void check_klass_subtype_slow_path(Register sub_klass,
619 Register super_klass,
620 Register temp_reg,
621 Register temp2_reg,
622 Label* L_success,
623 Label* L_failure,
624 bool set_cond_codes = false);
625
626 // The 64-bit version, which may do a hashed subclass lookup.
627 void check_klass_subtype_slow_path(Register sub_klass,
628 Register super_klass,
629 Register temp_reg,
630 Register temp2_reg,
631 Register temp3_reg,
632 Register temp4_reg,
633 Label* L_success,
634 Label* L_failure);
635
636 // Three parts of a hashed subclass lookup: a simple linear search,
637 // a table lookup, and a fallback that does linear probing in the
638 // event of a hash collision.
639 void check_klass_subtype_slow_path_linear(Register sub_klass,
640 Register super_klass,
641 Register temp_reg,
642 Register temp2_reg,
643 Label* L_success,
644 Label* L_failure,
645 bool set_cond_codes = false);
646 void check_klass_subtype_slow_path_table(Register sub_klass,
647 Register super_klass,
648 Register temp_reg,
649 Register temp2_reg,
650 Register temp3_reg,
651 Register result_reg,
652 Label* L_success,
653 Label* L_failure);
654 void hashed_check_klass_subtype_slow_path(Register sub_klass,
655 Register super_klass,
656 Register temp_reg,
657 Label* L_success,
658 Label* L_failure);
659
660 // As above, but with a constant super_klass.
661 // The result is in Register result, not the condition codes.
662 void lookup_secondary_supers_table_const(Register sub_klass,
663 Register super_klass,
664 Register temp1,
665 Register temp2,
666 Register temp3,
667 Register temp4,
668 Register result,
669 u1 super_klass_slot);
670
671 using Assembler::salq;
672 void salq(Register dest, Register count);
673 using Assembler::rorq;
674 void rorq(Register dest, Register count);
675 void lookup_secondary_supers_table_var(Register sub_klass,
676 Register super_klass,
677 Register temp1,
678 Register temp2,
679 Register temp3,
680 Register temp4,
681 Register result);
682
683 void lookup_secondary_supers_table_slow_path(Register r_super_klass,
684 Register r_array_base,
685 Register r_array_index,
686 Register r_bitmap,
687 Register temp1,
688 Register temp2,
689 Label* L_success,
690 Label* L_failure = nullptr);
691
692 void verify_secondary_supers_table(Register r_sub_klass,
693 Register r_super_klass,
694 Register expected,
695 Register temp1,
696 Register temp2,
697 Register temp3);
698
699 void repne_scanq(Register addr, Register value, Register count, Register limit,
700 Label* L_success,
701 Label* L_failure = nullptr);
702
703 // If r is valid, return r.
704 // If r is invalid, remove a register r2 from available_regs, add r2
705 // to regs_to_push, then return r2.
706 Register allocate_if_noreg(const Register r,
707 RegSetIterator<Register> &available_regs,
708 RegSet ®s_to_push);
709
710 // Simplified, combined version, good for typical uses.
711 // Falls through on failure.
712 void check_klass_subtype(Register sub_klass,
713 Register super_klass,
714 Register temp_reg,
715 Label& L_success);
716
717 void clinit_barrier(Register klass,
718 Label* L_fast_path = nullptr,
719 Label* L_slow_path = nullptr);
720
721 // method handles (JSR 292)
722 Address argument_address(RegisterOrConstant arg_slot, int extra_slot_offset = 0);
723
724 // Debugging
725
726 // only if +VerifyOops
727 void _verify_oop(Register reg, const char* s, const char* file, int line);
728 void _verify_oop_addr(Address addr, const char* s, const char* file, int line);
729
730 void _verify_oop_checked(Register reg, const char* s, const char* file, int line) {
731 if (VerifyOops) {
732 _verify_oop(reg, s, file, line);
733 }
734 }
735 void _verify_oop_addr_checked(Address reg, const char* s, const char* file, int line) {
736 if (VerifyOops) {
737 _verify_oop_addr(reg, s, file, line);
738 }
739 }
740
741 // TODO: verify method and klass metadata (compare against vptr?)
742 void _verify_method_ptr(Register reg, const char * msg, const char * file, int line) {}
743 void _verify_klass_ptr(Register reg, const char * msg, const char * file, int line){}
744
745 #define verify_oop(reg) _verify_oop_checked(reg, "broken oop " #reg, __FILE__, __LINE__)
746 #define verify_oop_msg(reg, msg) _verify_oop_checked(reg, "broken oop " #reg ", " #msg, __FILE__, __LINE__)
747 #define verify_oop_addr(addr) _verify_oop_addr_checked(addr, "broken oop addr " #addr, __FILE__, __LINE__)
748 #define verify_method_ptr(reg) _verify_method_ptr(reg, "broken method " #reg, __FILE__, __LINE__)
749 #define verify_klass_ptr(reg) _verify_klass_ptr(reg, "broken klass " #reg, __FILE__, __LINE__)
750
751 // Verify or restore cpu control state after JNI call
752 void restore_cpu_control_state_after_jni(Register rscratch);
753
754 // prints msg, dumps registers and stops execution
755 void stop(const char* msg);
756
757 // prints msg and continues
758 void warn(const char* msg);
759
760 // dumps registers and other state
761 void print_state();
762
763 static void debug32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip, char* msg);
764 static void debug64(char* msg, int64_t pc, int64_t regs[]);
765 static void print_state32(int rdi, int rsi, int rbp, int rsp, int rbx, int rdx, int rcx, int rax, int eip);
766 static void print_state64(int64_t pc, int64_t regs[]);
767
768 void os_breakpoint();
769
770 void untested() { stop("untested"); }
771
772 void unimplemented(const char* what = "");
773
774 void should_not_reach_here() { stop("should not reach here"); }
775
776 void print_CPU_state();
777
778 // Stack overflow checking
779 void bang_stack_with_offset(int offset) {
780 // stack grows down, caller passes positive offset
781 assert(offset > 0, "must bang with negative offset");
782 movl(Address(rsp, (-offset)), rax);
783 }
784
785 // Writes to stack successive pages until offset reached to check for
786 // stack overflow + shadow pages. Also, clobbers tmp
787 void bang_stack_size(Register size, Register tmp);
788
789 // Check for reserved stack access in method being exited (for JIT)
790 void reserved_stack_check();
791
792 void safepoint_poll(Label& slow_path, bool at_return, bool in_nmethod);
793
794 void verify_tlab();
795
796 static Condition negate_condition(Condition cond);
797
798 // Instructions that use AddressLiteral operands. These instruction can handle 32bit/64bit
799 // operands. In general the names are modified to avoid hiding the instruction in Assembler
800 // so that we don't need to implement all the varieties in the Assembler with trivial wrappers
801 // here in MacroAssembler. The major exception to this rule is call
802
803 // Arithmetics
804
805
806 void addptr(Address dst, int32_t src) { addq(dst, src); }
807 void addptr(Address dst, Register src);
808
809 void addptr(Register dst, Address src) { addq(dst, src); }
810 void addptr(Register dst, int32_t src);
811 void addptr(Register dst, Register src);
812 void addptr(Register dst, RegisterOrConstant src) {
813 if (src.is_constant()) addptr(dst, checked_cast<int>(src.as_constant()));
814 else addptr(dst, src.as_register());
815 }
816
817 void andptr(Register dst, int32_t src);
818 void andptr(Register src1, Register src2) { andq(src1, src2); }
819 void andptr(Register dst, Address src) { andq(dst, src); }
820
821 using Assembler::andq;
822 void andq(Register dst, AddressLiteral src, Register rscratch = noreg);
823
824 void cmp8(AddressLiteral src1, int imm, Register rscratch = noreg);
825
826 // renamed to drag out the casting of address to int32_t/intptr_t
827 void cmp32(Register src1, int32_t imm);
828
829 void cmp32(AddressLiteral src1, int32_t imm, Register rscratch = noreg);
830 // compare reg - mem, or reg - &mem
831 void cmp32(Register src1, AddressLiteral src2, Register rscratch = noreg);
832
833 void cmp32(Register src1, Address src2);
834
835 void cmpoop(Register src1, Register src2);
836 void cmpoop(Register src1, Address src2);
837 void cmpoop(Register dst, jobject obj, Register rscratch);
838
839 // NOTE src2 must be the lval. This is NOT an mem-mem compare
840 void cmpptr(Address src1, AddressLiteral src2, Register rscratch);
841
842 void cmpptr(Register src1, AddressLiteral src2, Register rscratch = noreg);
843
844 void cmpptr(Register src1, Register src2) { cmpq(src1, src2); }
845 void cmpptr(Register src1, Address src2) { cmpq(src1, src2); }
846
847 void cmpptr(Register src1, int32_t src2) { cmpq(src1, src2); }
848 void cmpptr(Address src1, int32_t src2) { cmpq(src1, src2); }
849
850 // cmp64 to avoild hiding cmpq
851 void cmp64(Register src1, AddressLiteral src, Register rscratch = noreg);
852
853 void cmpxchgptr(Register reg, Address adr);
854
855 void locked_cmpxchgptr(Register reg, AddressLiteral adr, Register rscratch = noreg);
856
857 void imulptr(Register dst, Register src) { imulq(dst, src); }
858 void imulptr(Register dst, Register src, int imm32) { imulq(dst, src, imm32); }
859
860
861 void negptr(Register dst) { negq(dst); }
862
863 void notptr(Register dst) { notq(dst); }
864
865 void shlptr(Register dst, int32_t shift);
866 void shlptr(Register dst) { shlq(dst); }
867
868 void shrptr(Register dst, int32_t shift);
869 void shrptr(Register dst) { shrq(dst); }
870
871 void sarptr(Register dst) { sarq(dst); }
872 void sarptr(Register dst, int32_t src) { sarq(dst, src); }
873
874 void subptr(Address dst, int32_t src) { subq(dst, src); }
875
876 void subptr(Register dst, Address src) { subq(dst, src); }
877 void subptr(Register dst, int32_t src);
878 // Force generation of a 4 byte immediate value even if it fits into 8bit
879 void subptr_imm32(Register dst, int32_t src);
880 void subptr(Register dst, Register src);
881 void subptr(Register dst, RegisterOrConstant src) {
882 if (src.is_constant()) subptr(dst, (int) src.as_constant());
883 else subptr(dst, src.as_register());
884 }
885
886 void sbbptr(Address dst, int32_t src) { sbbq(dst, src); }
887 void sbbptr(Register dst, int32_t src) { sbbq(dst, src); }
888
889 void xchgptr(Register src1, Register src2) { xchgq(src1, src2); }
890 void xchgptr(Register src1, Address src2) { xchgq(src1, src2); }
891
892 void xaddptr(Address src1, Register src2) { xaddq(src1, src2); }
893
894
895
896 // Helper functions for statistics gathering.
897 // Conditionally (atomically, on MPs) increments passed counter address, preserving condition codes.
898 void cond_inc32(Condition cond, AddressLiteral counter_addr, Register rscratch = noreg);
899 // Unconditional atomic increment.
900 void atomic_incl(Address counter_addr);
901 void atomic_incl(AddressLiteral counter_addr, Register rscratch = noreg);
902 void atomic_incq(Address counter_addr);
903 void atomic_incq(AddressLiteral counter_addr, Register rscratch = noreg);
904 void atomic_incptr(AddressLiteral counter_addr, Register rscratch = noreg) { atomic_incq(counter_addr, rscratch); }
905 void atomic_incptr(Address counter_addr) { atomic_incq(counter_addr); }
906
907 using Assembler::lea;
908 void lea(Register dst, AddressLiteral adr);
909 void lea(Address dst, AddressLiteral adr, Register rscratch);
910
911 void leal32(Register dst, Address src) { leal(dst, src); }
912
913 // Import other testl() methods from the parent class or else
914 // they will be hidden by the following overriding declaration.
915 using Assembler::testl;
916 void testl(Address dst, int32_t imm32);
917 void testl(Register dst, int32_t imm32);
918 void testl(Register dst, AddressLiteral src); // requires reachable address
919 using Assembler::testq;
920 void testq(Address dst, int32_t imm32);
921 void testq(Register dst, int32_t imm32);
922
923 void orptr(Register dst, Address src) { orq(dst, src); }
924 void orptr(Register dst, Register src) { orq(dst, src); }
925 void orptr(Register dst, int32_t src) { orq(dst, src); }
926 void orptr(Address dst, int32_t imm32) { orq(dst, imm32); }
927
928 void testptr(Register src, int32_t imm32) { testq(src, imm32); }
929 void testptr(Register src1, Address src2) { testq(src1, src2); }
930 void testptr(Address src, int32_t imm32) { testq(src, imm32); }
931 void testptr(Register src1, Register src2);
932
933 void xorptr(Register dst, Register src) { xorq(dst, src); }
934 void xorptr(Register dst, Address src) { xorq(dst, src); }
935
936 // Calls
937
938 void call(Label& L, relocInfo::relocType rtype);
939 void call(Register entry);
940 void call(Address addr) { Assembler::call(addr); }
941
942 // NOTE: this call transfers to the effective address of entry NOT
943 // the address contained by entry. This is because this is more natural
944 // for jumps/calls.
945 void call(AddressLiteral entry, Register rscratch = rax);
946
947 // Emit the CompiledIC call idiom
948 void ic_call(address entry, jint method_index = 0);
949 static int ic_check_size();
950 int ic_check(int end_alignment);
951
952 void emit_static_call_stub();
953
954 // Jumps
955
956 // NOTE: these jumps transfer to the effective address of dst NOT
957 // the address contained by dst. This is because this is more natural
958 // for jumps/calls.
959 void jump(AddressLiteral dst, Register rscratch = noreg);
960
961 void jump_cc(Condition cc, AddressLiteral dst, Register rscratch = noreg);
962
963 // 32bit can do a case table jump in one instruction but we no longer allow the base
964 // to be installed in the Address class. This jump will transfer to the address
965 // contained in the location described by entry (not the address of entry)
966 void jump(ArrayAddress entry, Register rscratch);
967
968 // Adding more natural conditional jump instructions
969 void ALWAYSINLINE jo(Label& L, bool maybe_short = true) { jcc(Assembler::overflow, L, maybe_short); }
970 void ALWAYSINLINE jno(Label& L, bool maybe_short = true) { jcc(Assembler::noOverflow, L, maybe_short); }
971 void ALWAYSINLINE js(Label& L, bool maybe_short = true) { jcc(Assembler::negative, L, maybe_short); }
972 void ALWAYSINLINE jns(Label& L, bool maybe_short = true) { jcc(Assembler::positive, L, maybe_short); }
973 void ALWAYSINLINE je(Label& L, bool maybe_short = true) { jcc(Assembler::equal, L, maybe_short); }
974 void ALWAYSINLINE jz(Label& L, bool maybe_short = true) { jcc(Assembler::zero, L, maybe_short); }
975 void ALWAYSINLINE jne(Label& L, bool maybe_short = true) { jcc(Assembler::notEqual, L, maybe_short); }
976 void ALWAYSINLINE jnz(Label& L, bool maybe_short = true) { jcc(Assembler::notZero, L, maybe_short); }
977 void ALWAYSINLINE jb(Label& L, bool maybe_short = true) { jcc(Assembler::below, L, maybe_short); }
978 void ALWAYSINLINE jnae(Label& L, bool maybe_short = true) { jcc(Assembler::below, L, maybe_short); }
979 void ALWAYSINLINE jc(Label& L, bool maybe_short = true) { jcc(Assembler::carrySet, L, maybe_short); }
980 void ALWAYSINLINE jnb(Label& L, bool maybe_short = true) { jcc(Assembler::aboveEqual, L, maybe_short); }
981 void ALWAYSINLINE jae(Label& L, bool maybe_short = true) { jcc(Assembler::aboveEqual, L, maybe_short); }
982 void ALWAYSINLINE jnc(Label& L, bool maybe_short = true) { jcc(Assembler::carryClear, L, maybe_short); }
983 void ALWAYSINLINE jbe(Label& L, bool maybe_short = true) { jcc(Assembler::belowEqual, L, maybe_short); }
984 void ALWAYSINLINE jna(Label& L, bool maybe_short = true) { jcc(Assembler::belowEqual, L, maybe_short); }
985 void ALWAYSINLINE ja(Label& L, bool maybe_short = true) { jcc(Assembler::above, L, maybe_short); }
986 void ALWAYSINLINE jnbe(Label& L, bool maybe_short = true) { jcc(Assembler::above, L, maybe_short); }
987 void ALWAYSINLINE jl(Label& L, bool maybe_short = true) { jcc(Assembler::less, L, maybe_short); }
988 void ALWAYSINLINE jnge(Label& L, bool maybe_short = true) { jcc(Assembler::less, L, maybe_short); }
989 void ALWAYSINLINE jge(Label& L, bool maybe_short = true) { jcc(Assembler::greaterEqual, L, maybe_short); }
990 void ALWAYSINLINE jnl(Label& L, bool maybe_short = true) { jcc(Assembler::greaterEqual, L, maybe_short); }
991 void ALWAYSINLINE jle(Label& L, bool maybe_short = true) { jcc(Assembler::lessEqual, L, maybe_short); }
992 void ALWAYSINLINE jng(Label& L, bool maybe_short = true) { jcc(Assembler::lessEqual, L, maybe_short); }
993 void ALWAYSINLINE jg(Label& L, bool maybe_short = true) { jcc(Assembler::greater, L, maybe_short); }
994 void ALWAYSINLINE jnle(Label& L, bool maybe_short = true) { jcc(Assembler::greater, L, maybe_short); }
995 void ALWAYSINLINE jp(Label& L, bool maybe_short = true) { jcc(Assembler::parity, L, maybe_short); }
996 void ALWAYSINLINE jpe(Label& L, bool maybe_short = true) { jcc(Assembler::parity, L, maybe_short); }
997 void ALWAYSINLINE jnp(Label& L, bool maybe_short = true) { jcc(Assembler::noParity, L, maybe_short); }
998 void ALWAYSINLINE jpo(Label& L, bool maybe_short = true) { jcc(Assembler::noParity, L, maybe_short); }
999 // * No condition for this * void ALWAYSINLINE jcxz(Label& L, bool maybe_short = true) { jcc(Assembler::cxz, L, maybe_short); }
1000 // * No condition for this * void ALWAYSINLINE jecxz(Label& L, bool maybe_short = true) { jcc(Assembler::cxz, L, maybe_short); }
1001
1002 // Short versions of the above
1003 void ALWAYSINLINE jo_b(Label& L) { jccb(Assembler::overflow, L); }
1004 void ALWAYSINLINE jno_b(Label& L) { jccb(Assembler::noOverflow, L); }
1005 void ALWAYSINLINE js_b(Label& L) { jccb(Assembler::negative, L); }
1006 void ALWAYSINLINE jns_b(Label& L) { jccb(Assembler::positive, L); }
1007 void ALWAYSINLINE je_b(Label& L) { jccb(Assembler::equal, L); }
1008 void ALWAYSINLINE jz_b(Label& L) { jccb(Assembler::zero, L); }
1009 void ALWAYSINLINE jne_b(Label& L) { jccb(Assembler::notEqual, L); }
1010 void ALWAYSINLINE jnz_b(Label& L) { jccb(Assembler::notZero, L); }
1011 void ALWAYSINLINE jb_b(Label& L) { jccb(Assembler::below, L); }
1012 void ALWAYSINLINE jnae_b(Label& L) { jccb(Assembler::below, L); }
1013 void ALWAYSINLINE jc_b(Label& L) { jccb(Assembler::carrySet, L); }
1014 void ALWAYSINLINE jnb_b(Label& L) { jccb(Assembler::aboveEqual, L); }
1015 void ALWAYSINLINE jae_b(Label& L) { jccb(Assembler::aboveEqual, L); }
1016 void ALWAYSINLINE jnc_b(Label& L) { jccb(Assembler::carryClear, L); }
1017 void ALWAYSINLINE jbe_b(Label& L) { jccb(Assembler::belowEqual, L); }
1018 void ALWAYSINLINE jna_b(Label& L) { jccb(Assembler::belowEqual, L); }
1019 void ALWAYSINLINE ja_b(Label& L) { jccb(Assembler::above, L); }
1020 void ALWAYSINLINE jnbe_b(Label& L) { jccb(Assembler::above, L); }
1021 void ALWAYSINLINE jl_b(Label& L) { jccb(Assembler::less, L); }
1022 void ALWAYSINLINE jnge_b(Label& L) { jccb(Assembler::less, L); }
1023 void ALWAYSINLINE jge_b(Label& L) { jccb(Assembler::greaterEqual, L); }
1024 void ALWAYSINLINE jnl_b(Label& L) { jccb(Assembler::greaterEqual, L); }
1025 void ALWAYSINLINE jle_b(Label& L) { jccb(Assembler::lessEqual, L); }
1026 void ALWAYSINLINE jng_b(Label& L) { jccb(Assembler::lessEqual, L); }
1027 void ALWAYSINLINE jg_b(Label& L) { jccb(Assembler::greater, L); }
1028 void ALWAYSINLINE jnle_b(Label& L) { jccb(Assembler::greater, L); }
1029 void ALWAYSINLINE jp_b(Label& L) { jccb(Assembler::parity, L); }
1030 void ALWAYSINLINE jpe_b(Label& L) { jccb(Assembler::parity, L); }
1031 void ALWAYSINLINE jnp_b(Label& L) { jccb(Assembler::noParity, L); }
1032 void ALWAYSINLINE jpo_b(Label& L) { jccb(Assembler::noParity, L); }
1033 // * No condition for this * void ALWAYSINLINE jcxz_b(Label& L) { jccb(Assembler::cxz, L); }
1034 // * No condition for this * void ALWAYSINLINE jecxz_b(Label& L) { jccb(Assembler::cxz, L); }
1035
1036 // Floating
1037
1038 void push_f(XMMRegister r);
1039 void pop_f(XMMRegister r);
1040 void push_d(XMMRegister r);
1041 void pop_d(XMMRegister r);
1042
1043 void push_ppx(Register src);
1044 void pop_ppx(Register dst);
1045
1046 void andpd(XMMRegister dst, XMMRegister src) { Assembler::andpd(dst, src); }
1047 void andpd(XMMRegister dst, Address src) { Assembler::andpd(dst, src); }
1048 void andpd(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1049
1050 void andnpd(XMMRegister dst, XMMRegister src) { Assembler::andnpd(dst, src); }
1051
1052 void andps(XMMRegister dst, XMMRegister src) { Assembler::andps(dst, src); }
1053 void andps(XMMRegister dst, Address src) { Assembler::andps(dst, src); }
1054 void andps(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1055
1056 void comiss(XMMRegister dst, XMMRegister src) { Assembler::comiss(dst, src); }
1057 void comiss(XMMRegister dst, Address src) { Assembler::comiss(dst, src); }
1058 void comiss(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1059
1060 void comisd(XMMRegister dst, XMMRegister src) { Assembler::comisd(dst, src); }
1061 void comisd(XMMRegister dst, Address src) { Assembler::comisd(dst, src); }
1062 void comisd(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1063
1064 void orpd(XMMRegister dst, XMMRegister src) { Assembler::orpd(dst, src); }
1065
1066 void cmp32_mxcsr_std(Address mxcsr_save, Register tmp, Register rscratch = noreg);
1067 void ldmxcsr(Address src) { Assembler::ldmxcsr(src); }
1068 void ldmxcsr(AddressLiteral src, Register rscratch = noreg);
1069
1070 private:
1071 void sha256_AVX2_one_round_compute(
1072 Register reg_old_h,
1073 Register reg_a,
1074 Register reg_b,
1075 Register reg_c,
1076 Register reg_d,
1077 Register reg_e,
1078 Register reg_f,
1079 Register reg_g,
1080 Register reg_h,
1081 int iter);
1082 void sha256_AVX2_four_rounds_compute_first(int start);
1083 void sha256_AVX2_four_rounds_compute_last(int start);
1084 void sha256_AVX2_one_round_and_sched(
1085 XMMRegister xmm_0, /* == ymm4 on 0, 1, 2, 3 iterations, then rotate 4 registers left on 4, 8, 12 iterations */
1086 XMMRegister xmm_1, /* ymm5 */ /* full cycle is 16 iterations */
1087 XMMRegister xmm_2, /* ymm6 */
1088 XMMRegister xmm_3, /* ymm7 */
1089 Register reg_a, /* == eax on 0 iteration, then rotate 8 register right on each next iteration */
1090 Register reg_b, /* ebx */ /* full cycle is 8 iterations */
1091 Register reg_c, /* edi */
1092 Register reg_d, /* esi */
1093 Register reg_e, /* r8d */
1094 Register reg_f, /* r9d */
1095 Register reg_g, /* r10d */
1096 Register reg_h, /* r11d */
1097 int iter);
1098
1099 void addm(int disp, Register r1, Register r2);
1100
1101 void sha512_AVX2_one_round_compute(Register old_h, Register a, Register b, Register c, Register d,
1102 Register e, Register f, Register g, Register h, int iteration);
1103
1104 void sha512_AVX2_one_round_and_schedule(XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1105 Register a, Register b, Register c, Register d, Register e, Register f,
1106 Register g, Register h, int iteration);
1107
1108 void addmq(int disp, Register r1, Register r2);
1109 public:
1110 void sha256_AVX2(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0,
1111 XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4,
1112 Register buf, Register state, Register ofs, Register limit, Register rsp,
1113 bool multi_block, XMMRegister shuf_mask);
1114 void sha512_AVX2(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0,
1115 XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4,
1116 Register buf, Register state, Register ofs, Register limit, Register rsp, bool multi_block,
1117 XMMRegister shuf_mask);
1118 void sha512_update_ni_x1(Register arg_hash, Register arg_msg, Register ofs, Register limit, bool multi_block);
1119
1120 void fast_md5(Register buf, Address state, Address ofs, Address limit,
1121 bool multi_block);
1122
1123 void fast_sha1(XMMRegister abcd, XMMRegister e0, XMMRegister e1, XMMRegister msg0,
1124 XMMRegister msg1, XMMRegister msg2, XMMRegister msg3, XMMRegister shuf_mask,
1125 Register buf, Register state, Register ofs, Register limit, Register rsp,
1126 bool multi_block);
1127
1128 void fast_sha256(XMMRegister msg, XMMRegister state0, XMMRegister state1, XMMRegister msgtmp0,
1129 XMMRegister msgtmp1, XMMRegister msgtmp2, XMMRegister msgtmp3, XMMRegister msgtmp4,
1130 Register buf, Register state, Register ofs, Register limit, Register rsp,
1131 bool multi_block, XMMRegister shuf_mask);
1132
1133 void fast_exp(XMMRegister xmm0, XMMRegister xmm1, XMMRegister xmm2, XMMRegister xmm3,
1134 XMMRegister xmm4, XMMRegister xmm5, XMMRegister xmm6, XMMRegister xmm7,
1135 Register rax, Register rcx, Register rdx, Register tmp);
1136
1137 private:
1138
1139 // these are private because users should be doing movflt/movdbl
1140
1141 void movss(Address dst, XMMRegister src) { Assembler::movss(dst, src); }
1142 void movss(XMMRegister dst, XMMRegister src) { Assembler::movss(dst, src); }
1143 void movss(XMMRegister dst, Address src) { Assembler::movss(dst, src); }
1144 void movss(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1145
1146 void movlpd(XMMRegister dst, Address src) {Assembler::movlpd(dst, src); }
1147 void movlpd(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1148
1149 public:
1150
1151 void addsd(XMMRegister dst, XMMRegister src) { Assembler::addsd(dst, src); }
1152 void addsd(XMMRegister dst, Address src) { Assembler::addsd(dst, src); }
1153 void addsd(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1154
1155 void addss(XMMRegister dst, XMMRegister src) { Assembler::addss(dst, src); }
1156 void addss(XMMRegister dst, Address src) { Assembler::addss(dst, src); }
1157 void addss(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1158
1159 void addpd(XMMRegister dst, XMMRegister src) { Assembler::addpd(dst, src); }
1160 void addpd(XMMRegister dst, Address src) { Assembler::addpd(dst, src); }
1161 void addpd(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1162
1163 using Assembler::vbroadcasti128;
1164 void vbroadcasti128(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch = noreg);
1165
1166 using Assembler::vbroadcastsd;
1167 void vbroadcastsd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch = noreg);
1168
1169 using Assembler::vbroadcastss;
1170 void vbroadcastss(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch = noreg);
1171
1172 // Vector float blend
1173 void vblendvps(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg);
1174 void vblendvpd(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister mask, int vector_len, bool compute_mask = true, XMMRegister scratch = xnoreg);
1175
1176 void divsd(XMMRegister dst, XMMRegister src) { Assembler::divsd(dst, src); }
1177 void divsd(XMMRegister dst, Address src) { Assembler::divsd(dst, src); }
1178 void divsd(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1179
1180 void divss(XMMRegister dst, XMMRegister src) { Assembler::divss(dst, src); }
1181 void divss(XMMRegister dst, Address src) { Assembler::divss(dst, src); }
1182 void divss(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1183
1184 // Move Unaligned Double Quadword
1185 void movdqu(Address dst, XMMRegister src);
1186 void movdqu(XMMRegister dst, XMMRegister src);
1187 void movdqu(XMMRegister dst, Address src);
1188 void movdqu(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1189
1190 void kmovwl(Register dst, KRegister src) { Assembler::kmovwl(dst, src); }
1191 void kmovwl(Address dst, KRegister src) { Assembler::kmovwl(dst, src); }
1192 void kmovwl(KRegister dst, KRegister src) { Assembler::kmovwl(dst, src); }
1193 void kmovwl(KRegister dst, Register src) { Assembler::kmovwl(dst, src); }
1194 void kmovwl(KRegister dst, Address src) { Assembler::kmovwl(dst, src); }
1195 void kmovwl(KRegister dst, AddressLiteral src, Register rscratch = noreg);
1196
1197 void kmovql(KRegister dst, KRegister src) { Assembler::kmovql(dst, src); }
1198 void kmovql(KRegister dst, Register src) { Assembler::kmovql(dst, src); }
1199 void kmovql(Register dst, KRegister src) { Assembler::kmovql(dst, src); }
1200 void kmovql(KRegister dst, Address src) { Assembler::kmovql(dst, src); }
1201 void kmovql(Address dst, KRegister src) { Assembler::kmovql(dst, src); }
1202 void kmovql(KRegister dst, AddressLiteral src, Register rscratch = noreg);
1203
1204 // Safe move operation, lowers down to 16bit moves for targets supporting
1205 // AVX512F feature and 64bit moves for targets supporting AVX512BW feature.
1206 void kmov(Address dst, KRegister src);
1207 void kmov(KRegister dst, Address src);
1208 void kmov(KRegister dst, KRegister src);
1209 void kmov(Register dst, KRegister src);
1210 void kmov(KRegister dst, Register src);
1211
1212 using Assembler::movddup;
1213 void movddup(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1214
1215 using Assembler::vmovddup;
1216 void vmovddup(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch = noreg);
1217
1218 // AVX Unaligned forms
1219 void vmovdqu(Address dst, XMMRegister src);
1220 void vmovdqu(XMMRegister dst, Address src);
1221 void vmovdqu(XMMRegister dst, XMMRegister src);
1222 void vmovdqu(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1223 void vmovdqu(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch = noreg);
1224 void vmovdqu(XMMRegister dst, XMMRegister src, int vector_len);
1225 void vmovdqu(XMMRegister dst, Address src, int vector_len);
1226 void vmovdqu(Address dst, XMMRegister src, int vector_len);
1227
1228 // AVX Aligned forms
1229 using Assembler::vmovdqa;
1230 void vmovdqa(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1231 void vmovdqa(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch = noreg);
1232
1233 // AVX512 Unaligned
1234 void evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len);
1235 void evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len);
1236 void evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, XMMRegister src, bool merge, int vector_len);
1237
1238 void evmovdqub(XMMRegister dst, XMMRegister src, int vector_len) { Assembler::evmovdqub(dst, src, vector_len); }
1239 void evmovdqub(XMMRegister dst, Address src, int vector_len) { Assembler::evmovdqub(dst, src, vector_len); }
1240
1241 void evmovdqub(XMMRegister dst, KRegister mask, XMMRegister src, bool merge, int vector_len) {
1242 if (dst->encoding() != src->encoding() || mask != k0) {
1243 Assembler::evmovdqub(dst, mask, src, merge, vector_len);
1244 }
1245 }
1246 void evmovdqub(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdqub(dst, mask, src, merge, vector_len); }
1247 void evmovdqub(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len) { Assembler::evmovdqub(dst, mask, src, merge, vector_len); }
1248 void evmovdqub(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch = noreg);
1249
1250 void evmovdquw(XMMRegister dst, XMMRegister src, int vector_len) { Assembler::evmovdquw(dst, src, vector_len); }
1251 void evmovdquw(Address dst, XMMRegister src, int vector_len) { Assembler::evmovdquw(dst, src, vector_len); }
1252 void evmovdquw(XMMRegister dst, Address src, int vector_len) { Assembler::evmovdquw(dst, src, vector_len); }
1253
1254 void evmovdquw(XMMRegister dst, KRegister mask, XMMRegister src, bool merge, int vector_len) {
1255 if (dst->encoding() != src->encoding() || mask != k0) {
1256 Assembler::evmovdquw(dst, mask, src, merge, vector_len);
1257 }
1258 }
1259 void evmovdquw(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len) { Assembler::evmovdquw(dst, mask, src, merge, vector_len); }
1260 void evmovdquw(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdquw(dst, mask, src, merge, vector_len); }
1261 void evmovdquw(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch = noreg);
1262
1263 void evmovdqul(XMMRegister dst, XMMRegister src, int vector_len) {
1264 if (dst->encoding() != src->encoding()) {
1265 Assembler::evmovdqul(dst, src, vector_len);
1266 }
1267 }
1268 void evmovdqul(Address dst, XMMRegister src, int vector_len) { Assembler::evmovdqul(dst, src, vector_len); }
1269 void evmovdqul(XMMRegister dst, Address src, int vector_len) { Assembler::evmovdqul(dst, src, vector_len); }
1270
1271 void evmovdqul(XMMRegister dst, KRegister mask, XMMRegister src, bool merge, int vector_len) {
1272 if (dst->encoding() != src->encoding() || mask != k0) {
1273 Assembler::evmovdqul(dst, mask, src, merge, vector_len);
1274 }
1275 }
1276 void evmovdqul(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdqul(dst, mask, src, merge, vector_len); }
1277 void evmovdqul(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len) { Assembler::evmovdqul(dst, mask, src, merge, vector_len); }
1278 void evmovdqul(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch = noreg);
1279
1280 void evmovdquq(XMMRegister dst, XMMRegister src, int vector_len) {
1281 if (dst->encoding() != src->encoding()) {
1282 Assembler::evmovdquq(dst, src, vector_len);
1283 }
1284 }
1285 void evmovdquq(XMMRegister dst, Address src, int vector_len) { Assembler::evmovdquq(dst, src, vector_len); }
1286 void evmovdquq(Address dst, XMMRegister src, int vector_len) { Assembler::evmovdquq(dst, src, vector_len); }
1287 void evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch = noreg);
1288 void evmovdqaq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch = noreg);
1289
1290 void evmovdquq(XMMRegister dst, KRegister mask, XMMRegister src, bool merge, int vector_len) {
1291 if (dst->encoding() != src->encoding() || mask != k0) {
1292 Assembler::evmovdquq(dst, mask, src, merge, vector_len);
1293 }
1294 }
1295 void evmovdquq(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdquq(dst, mask, src, merge, vector_len); }
1296 void evmovdquq(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len) { Assembler::evmovdquq(dst, mask, src, merge, vector_len); }
1297 void evmovdquq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch = noreg);
1298 void evmovdqaq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register rscratch = noreg);
1299
1300 using Assembler::movapd;
1301 void movapd(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1302
1303 // Move Aligned Double Quadword
1304 void movdqa(XMMRegister dst, XMMRegister src) { Assembler::movdqa(dst, src); }
1305 void movdqa(XMMRegister dst, Address src) { Assembler::movdqa(dst, src); }
1306 void movdqa(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1307
1308 void movsd(Address dst, XMMRegister src) { Assembler::movsd(dst, src); }
1309 void movsd(XMMRegister dst, XMMRegister src) { Assembler::movsd(dst, src); }
1310 void movsd(XMMRegister dst, Address src) { Assembler::movsd(dst, src); }
1311 void movsd(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1312
1313 void mulpd(XMMRegister dst, XMMRegister src) { Assembler::mulpd(dst, src); }
1314 void mulpd(XMMRegister dst, Address src) { Assembler::mulpd(dst, src); }
1315 void mulpd(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1316
1317 void mulsd(XMMRegister dst, XMMRegister src) { Assembler::mulsd(dst, src); }
1318 void mulsd(XMMRegister dst, Address src) { Assembler::mulsd(dst, src); }
1319 void mulsd(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1320
1321 void mulss(XMMRegister dst, XMMRegister src) { Assembler::mulss(dst, src); }
1322 void mulss(XMMRegister dst, Address src) { Assembler::mulss(dst, src); }
1323 void mulss(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1324
1325 // Carry-Less Multiplication Quadword
1326 void pclmulldq(XMMRegister dst, XMMRegister src) {
1327 // 0x00 - multiply lower 64 bits [0:63]
1328 Assembler::pclmulqdq(dst, src, 0x00);
1329 }
1330 void pclmulhdq(XMMRegister dst, XMMRegister src) {
1331 // 0x11 - multiply upper 64 bits [64:127]
1332 Assembler::pclmulqdq(dst, src, 0x11);
1333 }
1334
1335 void pcmpeqb(XMMRegister dst, XMMRegister src);
1336 void pcmpeqw(XMMRegister dst, XMMRegister src);
1337
1338 void pcmpestri(XMMRegister dst, Address src, int imm8);
1339 void pcmpestri(XMMRegister dst, XMMRegister src, int imm8);
1340
1341 void pmovzxbw(XMMRegister dst, XMMRegister src);
1342 void pmovzxbw(XMMRegister dst, Address src);
1343
1344 void pmovmskb(Register dst, XMMRegister src);
1345
1346 void ptest(XMMRegister dst, XMMRegister src);
1347
1348 void roundsd(XMMRegister dst, XMMRegister src, int32_t rmode) { Assembler::roundsd(dst, src, rmode); }
1349 void roundsd(XMMRegister dst, Address src, int32_t rmode) { Assembler::roundsd(dst, src, rmode); }
1350 void roundsd(XMMRegister dst, AddressLiteral src, int32_t rmode, Register rscratch = noreg);
1351
1352 void sqrtss(XMMRegister dst, XMMRegister src) { Assembler::sqrtss(dst, src); }
1353 void sqrtss(XMMRegister dst, Address src) { Assembler::sqrtss(dst, src); }
1354 void sqrtss(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1355
1356 void subsd(XMMRegister dst, XMMRegister src) { Assembler::subsd(dst, src); }
1357 void subsd(XMMRegister dst, Address src) { Assembler::subsd(dst, src); }
1358 void subsd(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1359
1360 void subss(XMMRegister dst, XMMRegister src) { Assembler::subss(dst, src); }
1361 void subss(XMMRegister dst, Address src) { Assembler::subss(dst, src); }
1362 void subss(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1363
1364 void ucomiss(XMMRegister dst, XMMRegister src) { Assembler::ucomiss(dst, src); }
1365 void ucomiss(XMMRegister dst, Address src) { Assembler::ucomiss(dst, src); }
1366 void ucomiss(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1367
1368 void ucomisd(XMMRegister dst, XMMRegister src) { Assembler::ucomisd(dst, src); }
1369 void ucomisd(XMMRegister dst, Address src) { Assembler::ucomisd(dst, src); }
1370 void ucomisd(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1371
1372 // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values
1373 void xorpd(XMMRegister dst, XMMRegister src);
1374 void xorpd(XMMRegister dst, Address src) { Assembler::xorpd(dst, src); }
1375 void xorpd(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1376
1377 // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values
1378 void xorps(XMMRegister dst, XMMRegister src);
1379 void xorps(XMMRegister dst, Address src) { Assembler::xorps(dst, src); }
1380 void xorps(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1381
1382 // Shuffle Bytes
1383 void pshufb(XMMRegister dst, XMMRegister src) { Assembler::pshufb(dst, src); }
1384 void pshufb(XMMRegister dst, Address src) { Assembler::pshufb(dst, src); }
1385 void pshufb(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1386 // AVX 3-operands instructions
1387
1388 void vaddsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddsd(dst, nds, src); }
1389 void vaddsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vaddsd(dst, nds, src); }
1390 void vaddsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch = noreg);
1391
1392 void vaddss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vaddss(dst, nds, src); }
1393 void vaddss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vaddss(dst, nds, src); }
1394 void vaddss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch = noreg);
1395
1396 void vabsss(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch = noreg);
1397 void vabssd(XMMRegister dst, XMMRegister nds, XMMRegister src, AddressLiteral negate_field, int vector_len, Register rscratch = noreg);
1398
1399 void vpaddb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1400 void vpaddb(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1401 void vpaddb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch = noreg);
1402
1403 void vpaddw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1404 void vpaddw(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1405
1406 void vpaddd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vpaddd(dst, nds, src, vector_len); }
1407 void vpaddd(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vpaddd(dst, nds, src, vector_len); }
1408 void vpaddd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch = noreg);
1409
1410 void vpand(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vpand(dst, nds, src, vector_len); }
1411 void vpand(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vpand(dst, nds, src, vector_len); }
1412 void vpand(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch = noreg);
1413
1414 using Assembler::vpbroadcastd;
1415 void vpbroadcastd(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch = noreg);
1416
1417 using Assembler::vpbroadcastq;
1418 void vpbroadcastq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch = noreg);
1419
1420 void vpcmpeqb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1421 void vpcmpeqb(XMMRegister dst, XMMRegister src1, Address src2, int vector_len);
1422
1423 void vpcmpeqw(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1424 void vpcmpeqw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1425 void evpcmpeqd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch = noreg);
1426
1427 // Vector compares
1428 void evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, XMMRegister src, int comparison, bool is_signed, int vector_len) {
1429 Assembler::evpcmpd(kdst, mask, nds, src, comparison, is_signed, vector_len);
1430 }
1431 void evpcmpd(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int comparison, bool is_signed, int vector_len, Register rscratch = noreg);
1432
1433 void evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, XMMRegister src, int comparison, bool is_signed, int vector_len) {
1434 Assembler::evpcmpq(kdst, mask, nds, src, comparison, is_signed, vector_len);
1435 }
1436 void evpcmpq(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int comparison, bool is_signed, int vector_len, Register rscratch = noreg);
1437
1438 void evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, XMMRegister src, int comparison, bool is_signed, int vector_len) {
1439 Assembler::evpcmpb(kdst, mask, nds, src, comparison, is_signed, vector_len);
1440 }
1441 void evpcmpb(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int comparison, bool is_signed, int vector_len, Register rscratch = noreg);
1442
1443 void evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, XMMRegister src, int comparison, bool is_signed, int vector_len) {
1444 Assembler::evpcmpw(kdst, mask, nds, src, comparison, is_signed, vector_len);
1445 }
1446 void evpcmpw(KRegister kdst, KRegister mask, XMMRegister nds, AddressLiteral src, int comparison, bool is_signed, int vector_len, Register rscratch = noreg);
1447
1448 void evpbroadcast(BasicType type, XMMRegister dst, Register src, int vector_len);
1449
1450 // Emit comparison instruction for the specified comparison predicate.
1451 void vpcmpCCW(XMMRegister dst, XMMRegister nds, XMMRegister src, XMMRegister xtmp, ComparisonPredicate cond, Width width, int vector_len);
1452 void vpcmpCC(XMMRegister dst, XMMRegister nds, XMMRegister src, int cond_encoding, Width width, int vector_len);
1453
1454 void vpmovzxbw(XMMRegister dst, Address src, int vector_len);
1455 void vpmovzxbw(XMMRegister dst, XMMRegister src, int vector_len) { Assembler::vpmovzxbw(dst, src, vector_len); }
1456
1457 void vpmovmskb(Register dst, XMMRegister src, int vector_len = Assembler::AVX_256bit);
1458
1459 void vpmullw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1460 void vpmullw(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1461
1462 void vpmulld(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vpmulld(dst, nds, src, vector_len); }
1463 void vpmulld(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vpmulld(dst, nds, src, vector_len); }
1464 void vpmulld(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch = noreg);
1465
1466 void vpmuldq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vpmuldq(dst, nds, src, vector_len); }
1467
1468 void vpsubb(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1469 void vpsubb(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1470
1471 void vpsubw(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len);
1472 void vpsubw(XMMRegister dst, XMMRegister nds, Address src, int vector_len);
1473
1474 void vpsraw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len);
1475 void vpsraw(XMMRegister dst, XMMRegister nds, int shift, int vector_len);
1476
1477 void evpsrad(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len);
1478 void evpsrad(XMMRegister dst, XMMRegister nds, int shift, int vector_len);
1479
1480 void evpsraq(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len);
1481 void evpsraq(XMMRegister dst, XMMRegister nds, int shift, int vector_len);
1482
1483 using Assembler::evpsllw;
1484 void evpsllw(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len, bool is_varshift) {
1485 if (!is_varshift) {
1486 Assembler::evpsllw(dst, mask, nds, src, merge, vector_len);
1487 } else {
1488 Assembler::evpsllvw(dst, mask, nds, src, merge, vector_len);
1489 }
1490 }
1491 void evpslld(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len, bool is_varshift) {
1492 if (!is_varshift) {
1493 Assembler::evpslld(dst, mask, nds, src, merge, vector_len);
1494 } else {
1495 Assembler::evpsllvd(dst, mask, nds, src, merge, vector_len);
1496 }
1497 }
1498 void evpsllq(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len, bool is_varshift) {
1499 if (!is_varshift) {
1500 Assembler::evpsllq(dst, mask, nds, src, merge, vector_len);
1501 } else {
1502 Assembler::evpsllvq(dst, mask, nds, src, merge, vector_len);
1503 }
1504 }
1505 void evpsrlw(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len, bool is_varshift) {
1506 if (!is_varshift) {
1507 Assembler::evpsrlw(dst, mask, nds, src, merge, vector_len);
1508 } else {
1509 Assembler::evpsrlvw(dst, mask, nds, src, merge, vector_len);
1510 }
1511 }
1512 void evpsrld(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len, bool is_varshift) {
1513 if (!is_varshift) {
1514 Assembler::evpsrld(dst, mask, nds, src, merge, vector_len);
1515 } else {
1516 Assembler::evpsrlvd(dst, mask, nds, src, merge, vector_len);
1517 }
1518 }
1519
1520 using Assembler::evpsrlq;
1521 void evpsrlq(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len, bool is_varshift) {
1522 if (!is_varshift) {
1523 Assembler::evpsrlq(dst, mask, nds, src, merge, vector_len);
1524 } else {
1525 Assembler::evpsrlvq(dst, mask, nds, src, merge, vector_len);
1526 }
1527 }
1528 using Assembler::evpsraw;
1529 void evpsraw(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len, bool is_varshift) {
1530 if (!is_varshift) {
1531 Assembler::evpsraw(dst, mask, nds, src, merge, vector_len);
1532 } else {
1533 Assembler::evpsravw(dst, mask, nds, src, merge, vector_len);
1534 }
1535 }
1536 using Assembler::evpsrad;
1537 void evpsrad(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len, bool is_varshift) {
1538 if (!is_varshift) {
1539 Assembler::evpsrad(dst, mask, nds, src, merge, vector_len);
1540 } else {
1541 Assembler::evpsravd(dst, mask, nds, src, merge, vector_len);
1542 }
1543 }
1544 using Assembler::evpsraq;
1545 void evpsraq(XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len, bool is_varshift) {
1546 if (!is_varshift) {
1547 Assembler::evpsraq(dst, mask, nds, src, merge, vector_len);
1548 } else {
1549 Assembler::evpsravq(dst, mask, nds, src, merge, vector_len);
1550 }
1551 }
1552
1553 void evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len);
1554 void evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len);
1555 void evpmins(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len);
1556 void evpmaxs(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len);
1557
1558 void evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len);
1559 void evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len);
1560 void evpminu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len);
1561 void evpmaxu(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len);
1562
1563 void vpsrlw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len);
1564 void vpsrlw(XMMRegister dst, XMMRegister nds, int shift, int vector_len);
1565
1566 void vpsllw(XMMRegister dst, XMMRegister nds, XMMRegister shift, int vector_len);
1567 void vpsllw(XMMRegister dst, XMMRegister nds, int shift, int vector_len);
1568
1569 void vptest(XMMRegister dst, XMMRegister src);
1570 void vptest(XMMRegister dst, XMMRegister src, int vector_len) { Assembler::vptest(dst, src, vector_len); }
1571
1572 void punpcklbw(XMMRegister dst, XMMRegister src);
1573 void punpcklbw(XMMRegister dst, Address src) { Assembler::punpcklbw(dst, src); }
1574
1575 void pshufd(XMMRegister dst, Address src, int mode);
1576 void pshufd(XMMRegister dst, XMMRegister src, int mode) { Assembler::pshufd(dst, src, mode); }
1577
1578 void pshuflw(XMMRegister dst, XMMRegister src, int mode);
1579 void pshuflw(XMMRegister dst, Address src, int mode) { Assembler::pshuflw(dst, src, mode); }
1580
1581 void vandpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vandpd(dst, nds, src, vector_len); }
1582 void vandpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vandpd(dst, nds, src, vector_len); }
1583 void vandpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch = noreg);
1584
1585 void vandps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vandps(dst, nds, src, vector_len); }
1586 void vandps(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vandps(dst, nds, src, vector_len); }
1587 void vandps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch = noreg);
1588
1589 void evpord(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register rscratch = noreg);
1590
1591 void vdivsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivsd(dst, nds, src); }
1592 void vdivsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vdivsd(dst, nds, src); }
1593 void vdivsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch = noreg);
1594
1595 void vdivss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vdivss(dst, nds, src); }
1596 void vdivss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vdivss(dst, nds, src); }
1597 void vdivss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch = noreg);
1598
1599 void vmulsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulsd(dst, nds, src); }
1600 void vmulsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vmulsd(dst, nds, src); }
1601 void vmulsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch = noreg);
1602
1603 void vmulss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vmulss(dst, nds, src); }
1604 void vmulss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vmulss(dst, nds, src); }
1605 void vmulss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch = noreg);
1606
1607 void vsubsd(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubsd(dst, nds, src); }
1608 void vsubsd(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vsubsd(dst, nds, src); }
1609 void vsubsd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch = noreg);
1610
1611 void vsubss(XMMRegister dst, XMMRegister nds, XMMRegister src) { Assembler::vsubss(dst, nds, src); }
1612 void vsubss(XMMRegister dst, XMMRegister nds, Address src) { Assembler::vsubss(dst, nds, src); }
1613 void vsubss(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch = noreg);
1614
1615 void vnegatess(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch = noreg);
1616 void vnegatesd(XMMRegister dst, XMMRegister nds, AddressLiteral src, Register rscratch = noreg);
1617
1618 // AVX Vector instructions
1619
1620 void vxorpd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vxorpd(dst, nds, src, vector_len); }
1621 void vxorpd(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vxorpd(dst, nds, src, vector_len); }
1622 void vxorpd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch = noreg);
1623
1624 void vxorps(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vxorps(dst, nds, src, vector_len); }
1625 void vxorps(XMMRegister dst, XMMRegister nds, Address src, int vector_len) { Assembler::vxorps(dst, nds, src, vector_len); }
1626 void vxorps(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch = noreg);
1627
1628 void vpxor(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
1629 if (UseAVX > 1 || (vector_len < 1)) // vpxor 256 bit is available only in AVX2
1630 Assembler::vpxor(dst, nds, src, vector_len);
1631 else
1632 Assembler::vxorpd(dst, nds, src, vector_len);
1633 }
1634 void vpxor(XMMRegister dst, XMMRegister nds, Address src, int vector_len) {
1635 if (UseAVX > 1 || (vector_len < 1)) // vpxor 256 bit is available only in AVX2
1636 Assembler::vpxor(dst, nds, src, vector_len);
1637 else
1638 Assembler::vxorpd(dst, nds, src, vector_len);
1639 }
1640 void vpxor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch = noreg);
1641
1642 // Simple version for AVX2 256bit vectors
1643 void vpxor(XMMRegister dst, XMMRegister src) {
1644 assert(UseAVX >= 2, "Should be at least AVX2");
1645 Assembler::vpxor(dst, dst, src, AVX_256bit);
1646 }
1647 void vpxor(XMMRegister dst, Address src) {
1648 assert(UseAVX >= 2, "Should be at least AVX2");
1649 Assembler::vpxor(dst, dst, src, AVX_256bit);
1650 }
1651
1652 void vpermd(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) { Assembler::vpermd(dst, nds, src, vector_len); }
1653 void vpermd(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch = noreg);
1654
1655 void vinserti128(XMMRegister dst, XMMRegister nds, XMMRegister src, uint8_t imm8) {
1656 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1657 Assembler::vinserti32x4(dst, nds, src, imm8);
1658 } else if (UseAVX > 1) {
1659 // vinserti128 is available only in AVX2
1660 Assembler::vinserti128(dst, nds, src, imm8);
1661 } else {
1662 Assembler::vinsertf128(dst, nds, src, imm8);
1663 }
1664 }
1665
1666 void vinserti128(XMMRegister dst, XMMRegister nds, Address src, uint8_t imm8) {
1667 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1668 Assembler::vinserti32x4(dst, nds, src, imm8);
1669 } else if (UseAVX > 1) {
1670 // vinserti128 is available only in AVX2
1671 Assembler::vinserti128(dst, nds, src, imm8);
1672 } else {
1673 Assembler::vinsertf128(dst, nds, src, imm8);
1674 }
1675 }
1676
1677 void vextracti128(XMMRegister dst, XMMRegister src, uint8_t imm8) {
1678 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1679 Assembler::vextracti32x4(dst, src, imm8);
1680 } else if (UseAVX > 1) {
1681 // vextracti128 is available only in AVX2
1682 Assembler::vextracti128(dst, src, imm8);
1683 } else {
1684 Assembler::vextractf128(dst, src, imm8);
1685 }
1686 }
1687
1688 void vextracti128(Address dst, XMMRegister src, uint8_t imm8) {
1689 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1690 Assembler::vextracti32x4(dst, src, imm8);
1691 } else if (UseAVX > 1) {
1692 // vextracti128 is available only in AVX2
1693 Assembler::vextracti128(dst, src, imm8);
1694 } else {
1695 Assembler::vextractf128(dst, src, imm8);
1696 }
1697 }
1698
1699 // 128bit copy to/from high 128 bits of 256bit (YMM) vector registers
1700 void vinserti128_high(XMMRegister dst, XMMRegister src) {
1701 vinserti128(dst, dst, src, 1);
1702 }
1703 void vinserti128_high(XMMRegister dst, Address src) {
1704 vinserti128(dst, dst, src, 1);
1705 }
1706 void vextracti128_high(XMMRegister dst, XMMRegister src) {
1707 vextracti128(dst, src, 1);
1708 }
1709 void vextracti128_high(Address dst, XMMRegister src) {
1710 vextracti128(dst, src, 1);
1711 }
1712
1713 void vinsertf128_high(XMMRegister dst, XMMRegister src) {
1714 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1715 Assembler::vinsertf32x4(dst, dst, src, 1);
1716 } else {
1717 Assembler::vinsertf128(dst, dst, src, 1);
1718 }
1719 }
1720
1721 void vinsertf128_high(XMMRegister dst, Address src) {
1722 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1723 Assembler::vinsertf32x4(dst, dst, src, 1);
1724 } else {
1725 Assembler::vinsertf128(dst, dst, src, 1);
1726 }
1727 }
1728
1729 void vextractf128_high(XMMRegister dst, XMMRegister src) {
1730 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1731 Assembler::vextractf32x4(dst, src, 1);
1732 } else {
1733 Assembler::vextractf128(dst, src, 1);
1734 }
1735 }
1736
1737 void vextractf128_high(Address dst, XMMRegister src) {
1738 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1739 Assembler::vextractf32x4(dst, src, 1);
1740 } else {
1741 Assembler::vextractf128(dst, src, 1);
1742 }
1743 }
1744
1745 // 256bit copy to/from high 256 bits of 512bit (ZMM) vector registers
1746 void vinserti64x4_high(XMMRegister dst, XMMRegister src) {
1747 Assembler::vinserti64x4(dst, dst, src, 1);
1748 }
1749 void vinsertf64x4_high(XMMRegister dst, XMMRegister src) {
1750 Assembler::vinsertf64x4(dst, dst, src, 1);
1751 }
1752 void vextracti64x4_high(XMMRegister dst, XMMRegister src) {
1753 Assembler::vextracti64x4(dst, src, 1);
1754 }
1755 void vextractf64x4_high(XMMRegister dst, XMMRegister src) {
1756 Assembler::vextractf64x4(dst, src, 1);
1757 }
1758 void vextractf64x4_high(Address dst, XMMRegister src) {
1759 Assembler::vextractf64x4(dst, src, 1);
1760 }
1761 void vinsertf64x4_high(XMMRegister dst, Address src) {
1762 Assembler::vinsertf64x4(dst, dst, src, 1);
1763 }
1764
1765 // 128bit copy to/from low 128 bits of 256bit (YMM) vector registers
1766 void vinserti128_low(XMMRegister dst, XMMRegister src) {
1767 vinserti128(dst, dst, src, 0);
1768 }
1769 void vinserti128_low(XMMRegister dst, Address src) {
1770 vinserti128(dst, dst, src, 0);
1771 }
1772 void vextracti128_low(XMMRegister dst, XMMRegister src) {
1773 vextracti128(dst, src, 0);
1774 }
1775 void vextracti128_low(Address dst, XMMRegister src) {
1776 vextracti128(dst, src, 0);
1777 }
1778
1779 void vinsertf128_low(XMMRegister dst, XMMRegister src) {
1780 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1781 Assembler::vinsertf32x4(dst, dst, src, 0);
1782 } else {
1783 Assembler::vinsertf128(dst, dst, src, 0);
1784 }
1785 }
1786
1787 void vinsertf128_low(XMMRegister dst, Address src) {
1788 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1789 Assembler::vinsertf32x4(dst, dst, src, 0);
1790 } else {
1791 Assembler::vinsertf128(dst, dst, src, 0);
1792 }
1793 }
1794
1795 void vextractf128_low(XMMRegister dst, XMMRegister src) {
1796 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1797 Assembler::vextractf32x4(dst, src, 0);
1798 } else {
1799 Assembler::vextractf128(dst, src, 0);
1800 }
1801 }
1802
1803 void vextractf128_low(Address dst, XMMRegister src) {
1804 if (UseAVX > 2 && VM_Version::supports_avx512novl()) {
1805 Assembler::vextractf32x4(dst, src, 0);
1806 } else {
1807 Assembler::vextractf128(dst, src, 0);
1808 }
1809 }
1810
1811 // 256bit copy to/from low 256 bits of 512bit (ZMM) vector registers
1812 void vinserti64x4_low(XMMRegister dst, XMMRegister src) {
1813 Assembler::vinserti64x4(dst, dst, src, 0);
1814 }
1815 void vinsertf64x4_low(XMMRegister dst, XMMRegister src) {
1816 Assembler::vinsertf64x4(dst, dst, src, 0);
1817 }
1818 void vextracti64x4_low(XMMRegister dst, XMMRegister src) {
1819 Assembler::vextracti64x4(dst, src, 0);
1820 }
1821 void vextractf64x4_low(XMMRegister dst, XMMRegister src) {
1822 Assembler::vextractf64x4(dst, src, 0);
1823 }
1824 void vextractf64x4_low(Address dst, XMMRegister src) {
1825 Assembler::vextractf64x4(dst, src, 0);
1826 }
1827 void vinsertf64x4_low(XMMRegister dst, Address src) {
1828 Assembler::vinsertf64x4(dst, dst, src, 0);
1829 }
1830
1831 // Carry-Less Multiplication Quadword
1832 void vpclmulldq(XMMRegister dst, XMMRegister nds, XMMRegister src) {
1833 // 0x00 - multiply lower 64 bits [0:63]
1834 Assembler::vpclmulqdq(dst, nds, src, 0x00);
1835 }
1836 void vpclmulhdq(XMMRegister dst, XMMRegister nds, XMMRegister src) {
1837 // 0x11 - multiply upper 64 bits [64:127]
1838 Assembler::vpclmulqdq(dst, nds, src, 0x11);
1839 }
1840 void vpclmullqhqdq(XMMRegister dst, XMMRegister nds, XMMRegister src) {
1841 // 0x10 - multiply nds[0:63] and src[64:127]
1842 Assembler::vpclmulqdq(dst, nds, src, 0x10);
1843 }
1844 void vpclmulhqlqdq(XMMRegister dst, XMMRegister nds, XMMRegister src) {
1845 //0x01 - multiply nds[64:127] and src[0:63]
1846 Assembler::vpclmulqdq(dst, nds, src, 0x01);
1847 }
1848
1849 void evpclmulldq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
1850 // 0x00 - multiply lower 64 bits [0:63]
1851 Assembler::evpclmulqdq(dst, nds, src, 0x00, vector_len);
1852 }
1853 void evpclmulhdq(XMMRegister dst, XMMRegister nds, XMMRegister src, int vector_len) {
1854 // 0x11 - multiply upper 64 bits [64:127]
1855 Assembler::evpclmulqdq(dst, nds, src, 0x11, vector_len);
1856 }
1857
1858 // AVX-512 mask operations.
1859 void kand(BasicType etype, KRegister dst, KRegister src1, KRegister src2);
1860 void kor(BasicType type, KRegister dst, KRegister src1, KRegister src2);
1861 void knot(uint masklen, KRegister dst, KRegister src, KRegister ktmp = knoreg, Register rtmp = noreg);
1862 void kxor(BasicType type, KRegister dst, KRegister src1, KRegister src2);
1863 void kortest(uint masklen, KRegister src1, KRegister src2);
1864 void ktest(uint masklen, KRegister src1, KRegister src2);
1865
1866 void evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len);
1867 void evperm(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len);
1868
1869 void evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len);
1870 void evor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len);
1871
1872 void evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len);
1873 void evand(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len);
1874
1875 void evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, XMMRegister src, bool merge, int vector_len);
1876 void evxor(BasicType type, XMMRegister dst, KRegister mask, XMMRegister nds, Address src, bool merge, int vector_len);
1877
1878 void evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc);
1879 void evrold(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc);
1880 void evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src, int shift, bool merge, int vlen_enc);
1881 void evrord(BasicType type, XMMRegister dst, KRegister mask, XMMRegister src1, XMMRegister src2, bool merge, int vlen_enc);
1882
1883 using Assembler::evpandq;
1884 void evpandq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch = noreg);
1885
1886 using Assembler::evpaddq;
1887 void evpaddq(XMMRegister dst, KRegister mask, XMMRegister nds, AddressLiteral src, bool merge, int vector_len, Register rscratch = noreg);
1888
1889 using Assembler::evporq;
1890 void evporq(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch = noreg);
1891
1892 using Assembler::vpshufb;
1893 void vpshufb(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch = noreg);
1894
1895 using Assembler::vpor;
1896 void vpor(XMMRegister dst, XMMRegister nds, AddressLiteral src, int vector_len, Register rscratch = noreg);
1897
1898 using Assembler::vpternlogq;
1899 void vpternlogq(XMMRegister dst, int imm8, XMMRegister src2, AddressLiteral src3, int vector_len, Register rscratch = noreg);
1900
1901 void cmov32( Condition cc, Register dst, Address src);
1902 void cmov32( Condition cc, Register dst, Register src);
1903
1904 void cmov( Condition cc, Register dst, Register src) { cmovptr(cc, dst, src); }
1905
1906 void cmovptr(Condition cc, Register dst, Address src) { cmovq(cc, dst, src); }
1907 void cmovptr(Condition cc, Register dst, Register src) { cmovq(cc, dst, src); }
1908
1909 void movoop(Register dst, jobject obj);
1910 void movoop(Address dst, jobject obj, Register rscratch);
1911
1912 void mov_metadata(Register dst, Metadata* obj);
1913 void mov_metadata(Address dst, Metadata* obj, Register rscratch);
1914
1915 void movptr(Register dst, Register src);
1916 void movptr(Register dst, Address src);
1917 void movptr(Register dst, AddressLiteral src);
1918 void movptr(Register dst, ArrayAddress src);
1919 void movptr(Register dst, intptr_t src);
1920 void movptr(Address dst, Register src);
1921 void movptr(Address dst, int32_t imm);
1922 void movptr(Address dst, intptr_t src, Register rscratch);
1923 void movptr(ArrayAddress dst, Register src, Register rscratch);
1924
1925 void movptr(Register dst, RegisterOrConstant src) {
1926 if (src.is_constant()) movptr(dst, src.as_constant());
1927 else movptr(dst, src.as_register());
1928 }
1929
1930
1931 // to avoid hiding movl
1932 void mov32(Register dst, AddressLiteral src);
1933 void mov32(AddressLiteral dst, Register src, Register rscratch = noreg);
1934
1935 // Import other mov() methods from the parent class or else
1936 // they will be hidden by the following overriding declaration.
1937 using Assembler::movdl;
1938 void movdl(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1939
1940 using Assembler::movq;
1941 void movq(XMMRegister dst, AddressLiteral src, Register rscratch = noreg);
1942
1943 // Can push value or effective address
1944 void pushptr(AddressLiteral src, Register rscratch);
1945
1946 void pushptr(Address src) { pushq(src); }
1947 void popptr(Address src) { popq(src); }
1948
1949 void pushoop(jobject obj, Register rscratch);
1950 void pushklass(Metadata* obj, Register rscratch);
1951
1952 // sign extend as need a l to ptr sized element
1953 void movl2ptr(Register dst, Address src) { movslq(dst, src); }
1954 void movl2ptr(Register dst, Register src) { movslq(dst, src); }
1955
1956
1957 public:
1958 // Inline type specific methods
1959 #include "asm/macroAssembler_common.hpp"
1960
1961 int store_inline_type_fields_to_buf(ciInlineKlass* vk, bool from_interpreter = true);
1962 bool move_helper(VMReg from, VMReg to, BasicType bt, RegState reg_state[]);
1963 bool unpack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index,
1964 VMReg from, int& from_index, VMRegPair* to, int to_count, int& to_index,
1965 RegState reg_state[]);
1966 bool pack_inline_helper(const GrowableArray<SigEntry>* sig, int& sig_index, int vtarg_index,
1967 VMRegPair* from, int from_count, int& from_index, VMReg to,
1968 RegState reg_state[], Register val_array);
1969 int extend_stack_for_inline_args(int args_on_stack);
1970 void remove_frame(int initial_framesize, bool needs_stack_repair);
1971 VMReg spill_reg_for(VMReg reg);
1972
1973 // clear memory of size 'cnt' qwords, starting at 'base';
1974 // if 'is_large' is set, do not try to produce short loop
1975 void clear_mem(Register base, Register cnt, Register val, XMMRegister xtmp, bool is_large, bool word_copy_only, KRegister mask=knoreg);
1976
1977 // clear memory initialization sequence for constant size;
1978 void clear_mem(Register base, int cnt, Register rtmp, XMMRegister xtmp, KRegister mask=knoreg);
1979
1980 // clear memory of size 'cnt' qwords, starting at 'base' using XMM/YMM registers
1981 void xmm_clear_mem(Register base, Register cnt, Register rtmp, XMMRegister xtmp, KRegister mask=knoreg);
1982
1983 // Fill primitive arrays
1984 void generate_fill(BasicType t, bool aligned,
1985 Register to, Register value, Register count,
1986 Register rtmp, XMMRegister xtmp);
1987
1988 void encode_iso_array(Register src, Register dst, Register len,
1989 XMMRegister tmp1, XMMRegister tmp2, XMMRegister tmp3,
1990 XMMRegister tmp4, Register tmp5, Register result, bool ascii);
1991
1992 void add2_with_carry(Register dest_hi, Register dest_lo, Register src1, Register src2);
1993 void multiply_64_x_64_loop(Register x, Register xstart, Register x_xstart,
1994 Register y, Register y_idx, Register z,
1995 Register carry, Register product,
1996 Register idx, Register kdx);
1997 void multiply_add_128_x_128(Register x_xstart, Register y, Register z,
1998 Register yz_idx, Register idx,
1999 Register carry, Register product, int offset);
2000 void multiply_128_x_128_bmi2_loop(Register y, Register z,
2001 Register carry, Register carry2,
2002 Register idx, Register jdx,
2003 Register yz_idx1, Register yz_idx2,
2004 Register tmp, Register tmp3, Register tmp4);
2005 void multiply_128_x_128_loop(Register x_xstart, Register y, Register z,
2006 Register yz_idx, Register idx, Register jdx,
2007 Register carry, Register product,
2008 Register carry2);
2009 void multiply_to_len(Register x, Register xlen, Register y, Register ylen, Register z, Register tmp0,
2010 Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5);
2011 void square_rshift(Register x, Register len, Register z, Register tmp1, Register tmp3,
2012 Register tmp4, Register tmp5, Register rdxReg, Register raxReg);
2013 void multiply_add_64_bmi2(Register sum, Register op1, Register op2, Register carry,
2014 Register tmp2);
2015 void multiply_add_64(Register sum, Register op1, Register op2, Register carry,
2016 Register rdxReg, Register raxReg);
2017 void add_one_64(Register z, Register zlen, Register carry, Register tmp1);
2018 void lshift_by_1(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2,
2019 Register tmp3, Register tmp4);
2020 void square_to_len(Register x, Register len, Register z, Register zlen, Register tmp1, Register tmp2,
2021 Register tmp3, Register tmp4, Register tmp5, Register rdxReg, Register raxReg);
2022
2023 void mul_add_128_x_32_loop(Register out, Register in, Register offset, Register len, Register tmp1,
2024 Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register rdxReg,
2025 Register raxReg);
2026 void mul_add(Register out, Register in, Register offset, Register len, Register k, Register tmp1,
2027 Register tmp2, Register tmp3, Register tmp4, Register tmp5, Register rdxReg,
2028 Register raxReg);
2029 void vectorized_mismatch(Register obja, Register objb, Register length, Register log2_array_indxscale,
2030 Register result, Register tmp1, Register tmp2,
2031 XMMRegister vec1, XMMRegister vec2, XMMRegister vec3);
2032
2033 // CRC32 code for java.util.zip.CRC32::updateBytes() intrinsic.
2034 void update_byte_crc32(Register crc, Register val, Register table);
2035 void kernel_crc32(Register crc, Register buf, Register len, Register table, Register tmp);
2036
2037 void kernel_crc32_avx512(Register crc, Register buf, Register len, Register table, Register tmp1, Register tmp2);
2038 void kernel_crc32_avx512_256B(Register crc, Register buf, Register len, Register key, Register pos,
2039 Register tmp1, Register tmp2, Label& L_barrett, Label& L_16B_reduction_loop,
2040 Label& L_get_last_two_xmms, Label& L_128_done, Label& L_cleanup);
2041
2042 // CRC32C code for java.util.zip.CRC32C::updateBytes() intrinsic
2043 // Note on a naming convention:
2044 // Prefix w = register only used on a Westmere+ architecture
2045 // Prefix n = register only used on a Nehalem architecture
2046 void crc32c_ipl_alg4(Register in_out, uint32_t n,
2047 Register tmp1, Register tmp2, Register tmp3);
2048 void crc32c_pclmulqdq(XMMRegister w_xtmp1,
2049 Register in_out,
2050 uint32_t const_or_pre_comp_const_index, bool is_pclmulqdq_supported,
2051 XMMRegister w_xtmp2,
2052 Register tmp1,
2053 Register n_tmp2, Register n_tmp3);
2054 void crc32c_rec_alt2(uint32_t const_or_pre_comp_const_index_u1, uint32_t const_or_pre_comp_const_index_u2, bool is_pclmulqdq_supported, Register in_out, Register in1, Register in2,
2055 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
2056 Register tmp1, Register tmp2,
2057 Register n_tmp3);
2058 void crc32c_proc_chunk(uint32_t size, uint32_t const_or_pre_comp_const_index_u1, uint32_t const_or_pre_comp_const_index_u2, bool is_pclmulqdq_supported,
2059 Register in_out1, Register in_out2, Register in_out3,
2060 Register tmp1, Register tmp2, Register tmp3,
2061 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
2062 Register tmp4, Register tmp5,
2063 Register n_tmp6);
2064 void crc32c_ipl_alg2_alt2(Register in_out, Register in1, Register in2,
2065 Register tmp1, Register tmp2, Register tmp3,
2066 Register tmp4, Register tmp5, Register tmp6,
2067 XMMRegister w_xtmp1, XMMRegister w_xtmp2, XMMRegister w_xtmp3,
2068 bool is_pclmulqdq_supported);
2069 // Fold 128-bit data chunk
2070 void fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, int offset);
2071 void fold_128bit_crc32(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, XMMRegister xbuf);
2072 // Fold 512-bit data chunk
2073 void fold512bit_crc32_avx512(XMMRegister xcrc, XMMRegister xK, XMMRegister xtmp, Register buf, Register pos, int offset);
2074 // Fold 8-bit data
2075 void fold_8bit_crc32(Register crc, Register table, Register tmp);
2076 void fold_8bit_crc32(XMMRegister crc, Register table, XMMRegister xtmp, Register tmp);
2077
2078 // Compress char[] array to byte[].
2079 void char_array_compress(Register src, Register dst, Register len,
2080 XMMRegister tmp1, XMMRegister tmp2, XMMRegister tmp3,
2081 XMMRegister tmp4, Register tmp5, Register result,
2082 KRegister mask1 = knoreg, KRegister mask2 = knoreg);
2083
2084 // Inflate byte[] array to char[].
2085 void byte_array_inflate(Register src, Register dst, Register len,
2086 XMMRegister tmp1, Register tmp2, KRegister mask = knoreg);
2087
2088 void fill_masked(BasicType bt, Address dst, XMMRegister xmm, KRegister mask,
2089 Register length, Register temp, int vec_enc);
2090
2091 void fill64_masked(uint shift, Register dst, int disp,
2092 XMMRegister xmm, KRegister mask, Register length,
2093 Register temp, bool use64byteVector = false);
2094
2095 void fill32_masked(uint shift, Register dst, int disp,
2096 XMMRegister xmm, KRegister mask, Register length,
2097 Register temp);
2098
2099 void fill32(Address dst, XMMRegister xmm);
2100
2101 void fill32(Register dst, int disp, XMMRegister xmm);
2102
2103 void fill64(Address dst, XMMRegister xmm, bool use64byteVector = false);
2104
2105 void fill64(Register dst, int dis, XMMRegister xmm, bool use64byteVector = false);
2106
2107 void convert_f2i(Register dst, XMMRegister src);
2108 void convert_d2i(Register dst, XMMRegister src);
2109 void convert_f2l(Register dst, XMMRegister src);
2110 void convert_d2l(Register dst, XMMRegister src);
2111 void round_double(Register dst, XMMRegister src, Register rtmp, Register rcx);
2112 void round_float(Register dst, XMMRegister src, Register rtmp, Register rcx);
2113
2114 void cache_wb(Address line);
2115 void cache_wbsync(bool is_pre);
2116
2117 #ifdef COMPILER2_OR_JVMCI
2118 void generate_fill_avx3(BasicType type, Register to, Register value,
2119 Register count, Register rtmp, XMMRegister xtmp);
2120 #endif // COMPILER2_OR_JVMCI
2121
2122 void vallones(XMMRegister dst, int vector_len);
2123
2124 void check_stack_alignment(Register sp, const char* msg, unsigned bias = 0, Register tmp = noreg);
2125
2126 void lightweight_lock(Register basic_lock, Register obj, Register reg_rax, Register tmp, Label& slow);
2127 void lightweight_unlock(Register obj, Register reg_rax, Register tmp, Label& slow);
2128
2129 void save_legacy_gprs();
2130 void restore_legacy_gprs();
2131 void setcc(Assembler::Condition comparison, Register dst);
2132 };
2133
2134 #endif // CPU_X86_MACROASSEMBLER_X86_HPP